JackAKirk

> Hi @JackAKirk on this new machine can you now check the atomics for 00:01.0 `sudo lspci -s 00:01.0 -vv | grep AtomicOpsCap` > > I expect it will show...

> > Hi @JackAKirk on this new machine can you now check the atomics for 00:01.0 `sudo lspci -s 00:01.0 -vv | grep AtomicOpsCap` > > I expect it will...

> > I'm pretty sure this indicates that the card supports pcie atomics right? So in this case I don't think that can be the issue. > > Can you...

> > > I'm pretty sure this indicates that the card supports pcie atomics right? So in this case I don't think that can be the issue. > > >...

> > dmesg: read kernel buffer failed: Operation not permitted > > Can you try to run dmesg with sudo. Also, can you post the output of this: `grep flags...

> @JackAKirk there is no indication of missing pcie atomics from the logs as far as I can see. > > 1. Can you try to call printf from the...

> @JackAKirk printf is part of unit tests https://github.com/ROCm/hip-tests/tree/develop/catch/unit/printf and these are quite well tested on gfx1030. Although I do not know the frequency each particular card is being used...

> Hi @JackAKirk so to confirm, If you replace assert() with printf() in the vectoradd test does it go through or it still fail with a bus error. Yeah if...

> Hi @JackAKirk but at the same time the printf unit tests pass, which is interesting. I would recommend we start with the printf unit test as a reference. The...

> Hi @JackAKirk the example with the assert works for me. I think that with the current implementation both the assert and printf require a synchronization on the host side...